Detects up to 10 simultaneous notes per moment (chords, harmonies, bass + melody together)
using spectral peak-picking, then plays back only those detected notes as piano tones —
not the original song audio. On a full mixed song this is an approximation, not true
vocal isolation: it can't cleanly separate voice from instruments, so expect some
extra or missing notes, especially with drums or heavy distortion.
Everything runs locally in your browser — no upload to any server.
The "emphasize center-panned audio" option is a rough heuristic (mid/side channel
processing), not true vocal separation — it reduces hard-panned instruments but
won't cleanly remove centered bass, kick drums, or other centered instruments.
For clean vocal-only stems, use a dedicated separation tool (e.g. a Demucs- or
Spleeter-based service) before uploading here.